[caret-users] Caret Segmentation

Donna Dierker donna at brainvis.wustl.edu
Wed Jun 6 10:29:51 CDT 2007


This morning, I pinpointed the culprit.  As expected, the output format
is irrelevant.  Ventricle filling is irrelevant (but notoriously less
reliable on monkeys -- especially ones whose heads are tilted a tad chin

The real keys were peaks and using the hindbrain low threshold option.
Just getting the peaks right wasn't enough; some stubborn hindbrain
caused the high thresh hindbrain removal to threshold most of the white
matter out of the segmentation.  You'll need to edit out the cerebellum,

Try running your hemisphere with these peaks and settings:

* GMpeak=84; WMpeak=104
* eye disconnect off
* hindbrain low threshold

A capture of the result is attached.


On 06/05/2007 04:31 PM, Donna Dierker wrote:
> Hi John,
> No trouble at all replicating the problem.  I got the same little blob 
> near the midline.
> Your peaks were way off, so I cranked them up -- around 80 on the GM, 
> 100 on the WM.  These can probably be tweaked more optimally.  But my 
> first try still gave the bad result, so I'm not sure it wasn't really 
> one of the other variables I changed:
> * output format to AFNI
> * eye disconnect off
> * hindbrain low threshold
> * ventricles and everything below basic segmentation off
> Normally, I change one variable at a time, but I'm headed home in five 
> minutes.
> My first try at cranking up the peaks (80 and 95, I think) still gave 
> the bad result.  Debug showed it was failing somewhere between eye fat 
> disconnection and hindbrain removal.  The above combo with peaks 
> 80,100 looks like it will give a decent result, albeit with some 
> tedious cerebellum to remove.  This is the drawback of the low thresh 
> hindbrain removal:  Less thresholding out of cortex you want, but more 
> thresholding in of hindbrain you don't want.  The segmentation is 
> still running, but looks more promising.
> Donna
> On 06/05/2007 03:14 PM, John Arsenault wrote:
>> Hi Donna,
>>   Thanks so much for taking a look.  I checked it out and it doesn't
>> appear to think it is mistakingly the right hemisphere but maybe that is
>> hidden to me. I uploaded it as chip_leftHem.zip.  Thanks so much for
>> checking this out,
>>       John
>>> Hi John,
>>> I think I've seen something like this once before, when the spec file
>>> thought it was a right hem, but it was really a left hem (or vice 
>>> versa).
>>> Once you rule out that possibility, then zip up your directory,
>>> including cropped hem, params, and any other files you think may help
>>> and upload the zip file here:
>>> http://pulvinar.wustl.edu/cgi-bin/upload.cgi
>>> I'll have a look.
>>> Donna
>>> On 06/05/2007 12:06 PM, John Arsenault wrote:
>>>> Hello,
>>>>   I am attempting to segment a macaque left hemisphere to create
>>>> surfaces
>>>> for data representation using caret.  I am using a very nice .35mm
>>>> isotropic anatomical with intensity normalization but for some reason
>>>> the
>>>> resultant segmentation of surefit is a small blob of medial frontal
>>>> white
>>>> matter that really has no anatomical correlate and is about 1/4 of the
>>>> size of the normal fiducial surface.  This doesn't seem to be 
>>>> related to
>>>> the image intensity because gray matter segements very well at 76(the
>>>> value prescribed by surefit) and white matter as well segments very 
>>>> well
>>>> at 100.  Also it doesn't seem to be the fact that the image is of 
>>>> .35mm
>>>> instead of the 1mm suggested since I was able to nicely segment the
>>>> right
>>>> hemisphere.  Another possible problem is the fact that the image has
>>>> been
>>>> skull stripped and the eyes have been disconnected but even if I
>>>> deselect
>>>> this option in surefit I get no difference. I have tried various 
>>>> changes
>>>> to the surefit process to get the image to segment properly and I 
>>>> can't
>>>> seem to get anything other then this bad segmentation except for an
>>>> error
>>>> about in ability to find big mask.  Any possible suggestions 
>>>> indicative
>>>> of
>>>> these problems?  Thanks for your help,
>>>>                                                               John
>>> _______________________________________________
>>> caret-users mailing list
>>> caret-users at brainvis.wustl.edu
>>> http://pulvinar.wustl.edu/mailman/listinfo/caret-users
>> _______________________________________________
>> caret-users mailing list
>> caret-users at brainvis.wustl.edu
>> http://pulvinar.wustl.edu/mailman/listinfo/caret-users

-------------- next part --------------
A non-text attachment was scrubbed...
Name: GM84WM104.jpg
Type: image/jpeg
Size: 160114 bytes
Desc: not available
URL: <http://brainvis.wustl.edu/pipermail/caret-users/attachments/20070606/9088fab4/attachment-0001.jpg>

More information about the caret-users mailing list